A conventional apparatus for describing and retrieving an image object with its shapes and a method thereof are disclosed in International Publication Patent No. WO00/67151. The conventional retrieval method retrieves an image object by describing the shape of the image object. It is developed taking notice that although an image object is shown in three-dimensional in video, it is, in fact, formed of two-dimensional images projected on a plane.
The two-dimensional shape or outline of an image object depends on a point of view, an angle of view, camera, and parameters of an optical system. The fact that the image object has different outlines according to different viewpoints is used as a descriptor. To retrieve an image object in an image retrieval system, the user inputs queries by presenting the image object to the image retrieval system or in selecting a view of the image object.
Then, the image retrieval system outputs representations for the queried image object, compares the query image object with the image objects stored in a database, and outputs an image object most closely matched with the query image object. The above-mentioned patent is technology used in description and retrieving steps after the image object is produced in computer graphics (CG) or a three-dimensional image object having multi-view data. Therefore, it can be used as a substructure of the present invention in certain applications.
Also, Korean Patent Laid-Open No. 10-2002-0059951 suggests a method for retrieving an image based on combinations of color and texture, which improves performance in image retrieval by combining characteristics of colors and textures appropriately. By combining the retrieved images with color and texture data, images more appropriate for human visual sense can be obtained. Also, more precise retrieval is possible by retrieving the image according to areas. The cited Korean Patent uses color characteristics and texture as a standard for classifying images, and it is limited to classification and description of two-dimensional images. Therefore, it can be also used as a substructure of the present invention to precisely retrieve the three-dimensional images retrieved in the present invention.
Another prior art, International Publication Patent No. WO00/46695, discloses descriptors for a video sequence and an image retrieval system using the descriptors. In the patent, motions of camera, observer and observing devices in the frames of a video sequence are expressed with descriptors from a video indexing view point. That is, the forms of fixing, panning, tracking, tilting, zooming, rolling, and other motions are divided and directed by two components, which indicate two different directions. Their values are expressed in histogram corresponding to a predetermined size of displacement. Just as the Korean Patent No. 10-2001-0001130, this invention is a technology based on the characteristics of two-dimensional images.
Normally, two-dimensional images are classified based on color formats, screen ratio and screen size and, after encoding, they are described based on the encoding method. Currently, image standardization groups, such as the Moving Picture Experts Group and the Joint Photographics Experts Group (JPEG), categorize profiles and levels of images based on color format, screen ratio and screen size of an image and select images suitable for each application area.
However, three-dimensional images are what a concept of multi-view or space is added to conventional two-dimensional image or a concept of computer graphics. Three-dimensional images with an additional viewpoint give cubical effect and panorama images obtained by mosaicking images of sequential viewpoints. Three-dimensional images adopting the concept of space can provide images at different angles at the same time.
In short, three-dimensional images can be provided in various forms according to what concept is added to a two-dimensional image and how the image is processed and displayed. Therefore, a systematic three-dimensional image description method and an image retrieval method based on the description method are required necessarily to retrieve three-dimensional images effectively in various application areas and to convert and exchange the images suitably to each application area and display apparatus.
However, the image classification method and the image description and retrieval method of prior arts, which are described above, are used limitedly to two-dimensional images. A classification structure and descriptors that are effective to classify three-dimensional images are not suggested in any prior art yet.